AI in an Era of Stupidity

I said yesterday that we would explore the zeitgeist, befogged as it is by a pervasive AI miasma. Well, tomorrow has arrived today, so here we go.

Let me get something out of the way at the outset. I believe we are living in a counterintuitive, paradoxical era in which rapid technological advances, occurring at a blistering pace unprecedented in human history, are accompanied by a rank stupidity.

Unfortunately, the latter is particularly conspicuous in positions of political, industrial, commercial, and technological leadership. Worse, the stupidity is intentional, even willful. Some people take enormous pride in being stupid, in not having to think critically or to consider the implications of their decisions and actions. I’m here to remind them that stupidity always carries a punitive cost, paid in part by the stupid but more often by innocent bystanders who suffer for the intellectually lethargic sins of the stupid.

So, let’s get back to AI, which at the end of the day (and at the end of history) will remain a human creation, a technological artifact. It does not have its own autonomous consciousness or teleology, but is created for purposes defined by its human progenitors.

Given that its human owners and custodians are drawn by the market’s charms, the primary purpose of AI is to make money, preferably lots of it. Any other story is just window dressing and marketing spin. This is why hype precedes and accompanies every generation of epochal growth in the technology industry. People need to aggressively sell technology if they ever hope to maximize their return on investment. Technologies might be complex and highly specialized, but the human motivations behind their commercial existences are relatively simple and straightforward. The market is the forum where fear and greed engage in a perpetual tug of war.

Does it matter if the marketing story gets ahead of a technology’s capabilities? If you want to be ethical about it, yes. But those in hot pursuit of billions of dollars — or, if you’re Elon Musk, trillions of dollars — put their ethics in long-term storage, for an eventual reunion when the billionaires and trillionaires get to reinvent themselves as benevolent philanthropists.

Tall Tales Cast a Big Shadow

AI’s proponents have gotten far ahead of the technology’s capabilities. That’s not merely my opinion. Many others have reached a similar conclusion, including MIT economist and Nobel laureate Daron Acemoglu, recently interviewed in the pages of Fortune.

Acemoglu doesn’t pull his punches. In these benighted times of cowed dishonesty and slobbering sycophancy, it’s refreshing and heartening to find somebody willing to speak with unrestrained candor. Acemoglu posits that the growth and productivity claims made for AI are vastly overstated. Contrary to the wildly optimistic forecasts emanating from Silicon Valley (which has a vested interest in the game) and Wall Street (which also has an abiding interest in the financial outcome), Acemoglu estimates that AI will deliver approximately 0.55% in total factor productivity gains in the decade ahead. Furthermore, Acemoglu calculates that only about 5% of tasks will be profitably automated in the near term. Nobody in the vendor community will be keen to put those numbers on its marketing collateral. Modest claims don’t exactly get the pulse quickening or the breath hot with the anticipation of financial congress.

As for how much of the current AI palaver that Acemoglu regards as intellectually serious (as in credible), he puts the figure at what I consider a mildly generous 20%. A quote from the article:

“I find all of this discussion of capitalism so brainless,” Acemoglu told Fortune, insisting that we should be focused on the “enormous increase” in corporate power and monopoly instead. “That’s what we should be talking about. What we should be talking about is the displacement and unequalizing roles of AI.” When asked how much of the discourse he finds, in his words “brainless,” he barely paused. “About 80%,” he said. He clarified that the thinking is rather speculative or close to fictional, not stupid per se.

Unlike Acemoglu, I’m willing to call our epidemic of ersatz thinking crushingly stupid. We’re losing the plot here, folks, but we seem to be enjoying the lurid detour into vapidity. It’s easier to go with the feculent flow than to truly think for oneself. In a sense, isn’t one of AI promises that you might not have to think for yourself, that you can delegate all that exhausting and tiresome thinking to automated machinery. You’ll be free to do . . . well, what exactly? That, of course, opens up a different door, to another room of discourse, one you’re free to contemplate at your leisure.

Acemoglu’s assessments of AI’s productivity potential are predicated on a methodology and framework that he has used to measure quantifiable gains during decades of automation. Quote:

“Productivity gains from automation, he explained, only materialize if machines can do tasks significantly cheaper or better than humans. If the improvement is marginal, or if integration costs eat into gains, the math doesn’t add up — even if the automation is widespread. “It’s not that you cannot get big productivity gains from automation,” he said. “It is that it’s not as easy as sometimes it’s presumed.”

Contingency: AI’s Achilles’ Heel

Later in the interview, Acemoglu suggests that the productivity gains forecast by Silicon Valley and Wall Street are not achievable by LLMs of any size, but would require “artificial general intelligence,” a term that, I believe, is fraught with ambiguity and inherent imprecision. Even the last word, “intelligence,” is perhaps subject to irresolvable debate.

Some of AI most passionate proponents will attempt to persuade you that LLMs are just a hop, skip, and a jump from attaining something approximating artificial general intelligence. Acemoglu is not buying that argument. He contends that current LLMs perform poorly in too many “dimensions of real-world work — they can’t read a room, they can’t connect non-obvious dots across domains, and they fail precisely where human judgment is most valuable.”

That’s a valid point, one I feel compelled to expand upon. In all our lives, professional and personal, contingency is an ever-present factor. External events — often uncontrollable, unpredictable, seemingly random — encroach and impinge on our plans and prescriptions. Contingency occurs with alarming regularity, every day of our lives. How can any model account for these developments, much less respond to them dynamically, prudently, reasonably, and, yes, intelligently? I think we’re a long distance from AGI being able to answer the insistent call of remorseless contingency.

At the end of the interview, Acemoglu considers what we, as a society, should do with AI. It’s a discussion of the human purpose of AI, of its human-defined teleology. The purpose of AI should not be limited to strictly technical horizons or to the financial gains of hyperscalers and AI billionaires. We should think about the bigger picture, about broader societal benefits, and — in my view — a reunion of enlightenment with self-interest. Anything else, as Acemoglu notes, would be a failure of intellect and imagination.